lable at ScienceDirect

Environmental Modelling & Software 82 (2016) 7e20

Contents lists avai

Environmental Modelling & Software

journal homepage: www.elsevier .com/locate/envsoft

Methods for translating narrative scenarios into quantitativeassessments of land use change

Varun Rao Mallampalli a, Georgia Mavrommati b, Jonathan Thompson c,Matthew Duveneck c, Spencer Meyer d, Arika Ligmann-Zielinska e,Caroline Gottschalk Druschke f, Kristen Hychka f, Melissa A. Kenney g, Kasper Kok h,Mark E. Borsuk a, *

a Thayer School of Engineering, Dartmouth College, Hanover, NH, USAb Environmental Studies Program, Dartmouth College, Hanover, NH, USAc Harvard Forest, Harvard University, Petersham, MA, USAd Yale School of Forestry and Management, New Haven, CT, USAe Department of Geography and Environmental Science and Policy Program, Michigan State University, East Lansing, MI, USAf Department of Natural Resources Science, University of Rhode Island, Kingston, RI, USAg Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, USAh Environmental Sciences, Wageningen University and Research Centre, Wageningen, Netherlands

a r t i c l e i n f o

Article history:Received 1 July 2015Received in revised form23 March 2016Accepted 11 April 2016

Keywords:Socio-ecological scenariosParticipatory processAgent-based modelFuzzy cognitive mapBayesian networkSystem dynamics

* Corresponding author.E-mail address: mark.borsuk@dartmouth.edu (M.E

http://dx.doi.org/10.1016/j.envsoft.2016.04.0111364-8152/© 2016 Elsevier Ltd. All rights reserved.

a b s t r a c t

In the land use and land cover (LULC) literature, narrative scenarios are qualitative descriptions ofplausible futures associated with a combination of socio-economic, policy, technological, and climatechanges. LULC models are then often used to translate these narrative descriptions into quantitativecharacterizations of possible future societal and ecological impacts and conditions. To respect the intentof the underlying scenario descriptions, this process of translation needs to be thoughtful, transparent,and reproducible. This paper evaluates the current state of the art in scenario translation methods andoutlines their relative advantages and disadvantages, as well as the respective roles of stakeholders andsubject matter experts. We summarize our findings in the form of a decision matrix that can assist landuse planners, scientists, and modelers in choosing a translation method appropriate to their situation.

© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Anticipating environmental impacts associated with future landuse is an important challenge within regional global change sci-ence. However, because land use is a product of complex socio-ecological factors, accurately predicting the drivers of land-usechange, let alone future land-use patterns, is typically impossible.Therefore, analyses of future land use often take a scenarioapproach, wherein several potential pathways are examinedwithout an attempt to make precise or probabilistic predictions;rather, a range of possibilities are considered. In this sense, sce-narios are “coherent and plausible stories, told in words and

. Borsuk).

numbers, about the possible co-evolutionary pathways of com-bined human and environmental systems” (Swart et al., 2004). Thecreation and analysis of land-use scenarios allows practitioners tointegrate diverse modes of knowledge and to explicitly recognizethose components of complex systems that are uncertain(Thompson et al., 2012). Consequently, land use scenarios are aubiquitous component of integrated environmental assessments atglobal (Alcamo, 2009), regional (Sleeter et al., 2012), and local(Carpenter et al., 2015) scales.

Scenarios takemany forms and varywidely in terms of how theyare developed. One useful distinction is between qualitative andquantitative scenarios (Alcamo, 2009). Qualitative land-use sce-narios are non-numerical descriptions of the way the future mayunfold depicted as narrative texts or storylines. Done well, quali-tative scenarios offer a compelling vision that showcases themyriad consequences and interdependencies of alternative land-

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e208

use decisions. Developing qualitative scenarios can be useful forgenerating ideas and strategies and for incorporating multipleviewpoints without requiring specific technical expertise. As such,the process of constructing scenarios can bridge gaps between andamong experts, decision-makers, and stakeholders (Welp et al.,2006). Participatory scenario development is increasingly used toensure that a range of viewpoints and expertise is captured byqualitative storylines and to maximize the legitimacy and salienceof the research for decision makers (Cash et al., 2002; Seppelt et al.,2011).

Quantitative land-use scenarios, in contrast, describe plausiblefutures using numerical descriptions of one or more of the rates,types, and spatial allocations of land uses associated with a po-tential pathway. Quantitative scenarios are typically designed tosupport numerical assessment of specific impacts. In many cases,quantitative scenarios are derived using spatially explicit simula-tion models of land use and land-cover change, often coupled tomodels of ecosystems, hydrology, or other affected components ofthe environment. Accordingly, the information required for thequantitative scenario is often specific and technical, thus confiningthe process to scientific experts. This runs the risk of excluding non-technical viewpoints and can decrease the accessibility of results tonon-experts (Voinov and Bousquet, 2010).

Often the value of scenario planning can be maximized whenthe inclusivity and creativity of qualitative scenarios is coupledwith the specificity of quantitative modeling. Indeed, coupledqualitative and quantitative scenarios are the basis of manyprominent scenario assessments in domains not directly related toland use assessment, such as the IPCC Special Report on EmissionScenarios (Nakicenovic and Swart, 2000) and the MillenniumEcosystem Assessment (Millennium Ecosystem Assessment, 2003).To date, the story and simulation (SAS) approach has been mostcommonly used to couple qualitative and quantitative scenarios. Inthis approach, scenarios are first defined by experts and/or stake-holders and subsequently translated into quantitative parametersthat feed into simulation models (Alcamo et al., 2008; Houet et al.,2016). As shown in Fig. 1, this may be an iterative process such thatnarrative scenarios are translated into quantitative models that arethen used to revise or enrich storylines based on the particularsimulation results.

Ultimately, any modeling approach used requires a “translation”step to relate qualitative narrative scenarios to specific simulationmodel formulations and input values. This translation needs toadhere to the assumptions and interactions contained in thenarrative, while also being compatible with the level of complexityof the chosenmodel (Kok, 2009). In this paper, we review the state-of-the-art of methods used to perform such translation for futureland-use scenarios. Our effort was initiated as part of a recentworkshop organized by the Scenarios, Services, and SocietyResearch Coordination Network (S3 RCN; s3rcn.org) funded by the

Fig. 1. Scenario development process. Dashed arrow indicates that scenario develop-ment can be an iterative process. The present paper addresses mainly the narrativetranslation step (gray box).

U.S. National Science Foundation. Based on a review of the land useand land change (LULC) scenario literature, we outline the relativeadvantages and disadvantages of the various translations methods.Many of thesemethods can also be used in the original constructionof the narrative scenario and/or in the model-based scenarioquantification. Finally, we provide a decision matrix summarizingour findings that can assist land use planners, scientists, andmodelers in choosing a translation method appropriate to theirsituation.

2. Translation methods

We review ten methods that can be used to translate betweennarrative scenario descriptions and quantitative simulationmodels.We evaluate each method according to the underlying conceptualfoundation, treatment of uncertainty, potential to accommodatestakeholder participation, relative level of resources required, andcompatibility with common simulation models. Table 1 provides asummary, including references to example applications. We alsoevaluate the relative advantages and disadvantages of each methodin particular applications. These criteria are further elaborated inthe discussion section and are summarized in the form of a decisionmatrix (Table 2). The purpose of this matrix is to assist scenariodevelopers andmodelers in the choice of an appropriate translationmethod for their particular situation.

2.1. System dynamics

2.1.1. Use in translationSystem dynamics (SD) is a method for framing and disen-

tangling the non-linear behavior of complex systems over time(Schmitt Olabisi et al., 2010). SD acknowledges that mental modelsare typically unable to capture features of complex systems such asfeedback loops, time delays and policy resistance (Sterman, 2012).SD uses graphical causal loop diagrams (Ford, 1999) to representcause and effect relationships and feedbacks. When used torepresent human-environment interactions, the SDmethod has thepotential to engage stakeholders in the process of understandingdynamics and defining scenarios (Mavrommati et al., 2014; SchmittOlabisi et al., 2010; Yu et al., 2011).

The SD approach can be used for qualitative-to-quantitativescenario translation either directly or indirectly. In the directmethod, researchers and stakeholders work together to understandthe system under study by: (i) identifying relationships among thesystem's key components and leverage points and then (ii) jointlyexploring plausible scenarios. This SAS-type approach typically firstyields an initial causal loop diagram corresponding to a business-as-usual scenario, which corresponds to a future scenario if nochanges are made to the system. This process can then be used togenerate additional scenarios by exploring relevant modifications(Mavrommati et al., 2014).

In the indirect method, the scenario development process pre-cedes the development of a causal loop diagram. For example, theMinnesota 2050 project (Schmitt Olabisi, Kapuscinski et al., 2010)used scenario visioning and stakeholder participatory modeling todevelop narrative scenarios describing cause-effect relationships.Researchers then used these qualitative scenarios to identify thesystem variables, linkages, and underlying assumptions necessaryto create a causal loop diagram (Fig. 2). Stakeholders then had theopportunity to provide their comments on the diagram at a followup meeting. Numerical implementation of a causal loop diagramrequires conversion to a stock-flow model along with additionalparameterization. In the case of the Minnesota 2050 project,empirical datawere used to parameterize themodel with a range ofvalues employed to explore various scenarios.

Table 1Summary of translation methods.

Method Conceptualfoundation

Treatment ofuncertainty

Potential forstakeholderparticipation

Level of resources/time required

Compatibility withsimulation models

Example references

System dynamics(Causal loopdiagrams)

Systems' thinking:mapping positive andnegative feedbackloops between systemparameters

Sensitivity analysis Very suitable forstakeholderengagement, withparticipationfacilitated throughworkshops.

An iterative processwith the level ofresources and timeinvolvementdepending on thelevel of stakeholders'participation and theresearch objective.

Causal loop diagramsare often used as abasis for buildingstock-flow models.Stand-alone CLDs canalso be used tounderstand leveragepoints for policyintervention.

(Mavrommati et al.,2014; Schmitt Olabisiet al., 2010; Sterman,2000)

Agent-based modeling Exploratory tool tounderstand the effectsof micro-level agentbehaviors on macro-level phenomenon

Sensitivity analysis orMonte Carlosimulation

Suitable forstakeholderengagement, withparticipationfacilitated throughworkshops.

An iterative processwith the level ofresources and timeinvolvementdepending on thelevel of stakeholders'participation and theresearch objective.

Agent based causaldiagrams are used as abasis for constructingagent based models.Stand-alone causaldiagrams can also beused to understandleverage points forpolicy intervention

(Janssen, 2005; Parkeret al., 2003)

Fuzzy cognitive maps Fuzzy logic; mappingfeedback loopsrepresenting strengthof a causalrelationship betweensystem variables

The treatment ofuncertainty isinherent. Diverse oruncertain judgmentsabout causalrelationships arerepresented as a fuzzyset or membershipfunction

Very suitable forstakeholderengagement, withparticipationfacilitated throughworkshops.

An iterative processwith the level ofresources and timeinvolvementdepending on thelevel of stakeholders'participation and theresearch objective.

The FCM itself can besimulated afterstrength of causalrelationships areidentified.

(Patel et al.,2007; vanVliet et al., 2010)

Fuzzy sets Fuzzy logic; variablesare represented as afuzzy view of diversejudgments

The treatment ofuncertainty isinherent. Diverse oruncertain judgmentsabout variables arerepresented as a fuzzyset or membershipfunction

Very suitable forstakeholderengagement inmultiple forms. It is aneasy method to teachto stakeholders

The method is simpleand multipleiterations can beexecuted in a singleworkshop.

The fuzzy numericaloutput can be useddirectly as inputs tostochastic models ordefuzzified for non-stochastic models.

(Alcamo, 2009;Dubrovsky et al.,2011; Kok et al., 2015)

Pairwise comparison Pairwise comparisonof storylines followedby translation ofqualitativeevaluations intonumerical values.

Sensitivity analysisbased on the diversityof expert judgments

Suitable forstakeholderengagement, withparticipationfacilitated throughworkshops.

Analytic hierarchyprocess is timeconsuming and needsseveral iterationsbefore evaluationsacross all expertsbecome consistent.

Provides direct inputfor a simulationmodel.

(Abildtrup et al., 2006;McDaniels et al., 2012)

Bayesian reasoning Bayesian reasoningthrough specificationof a prior distribution(based on a baselinescenario) andlikelihood distribution(based on plausiblescenario).

The treatment ofuncertainty isinherent. The inputsand outputs are bothprobabilitydistributions.

It requires somedegree of specializedknowledge foreffective stakeholderparticipation and tofind innovative waysto do so.

Depends on theobjective and numberof experts involved inthe process.Deliberativeassessments amongexperts can make theprocess time-consuming

Provides direct inputto stochasticsimulation models.

(Kemp-Benedict,2013)

Bayesian networks Influence diagramsmap dependenceamong variables.Conditionalprobabilities quantifythe strength of thedependences.

The treatment ofuncertainty isinherent; conditionalprobabilitydistributions

This method issuitable forstakeholderengagement, but itrequires stakeholdersto understand theconcept of conditionalprobabilities.

Depends on the scaleand objective of thestudy. It can be a time-intensive process ifconditionalprobabilities arecontinuously updatedbased on newevidence.

Influence diagramsand CPTs lead tofunctional BayesianNetworks, whichthemselves may leadto inputs for othermodels.

(Aalders, 2008;Marcot et al., 2006)

Role playing games A combinedrepresentation ofqualitative storylinesand the local decisionmaking processes inthe form of a game

Not generallyapplicable.

Stakeholderengagement isessential for thismethod, whoseparticipation isfacilitated throughworkshops.

Depends on thenumber ofstakeholders involvedand the objective ofthe study. Consensusand trust buildingexercises can be timeconsuming andresource intensive

Stakeholders'decisions during thegame can serve asinputs to simulationmodels. Stand-aloneRPGs can also producequantitative maps ofland use change.

(Castella et al., 2005;Lamarque et al., 2013)

Depends on the wayparameters are

Depends on the waythe survey is

Provides direct inputsto a simulation model.

(McDaniels et al.,2012)

(continued on next page)

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 9

Table 1 (continued )

Method Conceptualfoundation

Treatment ofuncertainty

Potential forstakeholderparticipation

Level of resources/time required

Compatibility withsimulation models

Example references

Surveys or semi-structuredinterviews

A survey of subjectivebeliefs of experts orstakeholders.

elicited. Uncertaintycan be inherent ifelicitations areprobabilistic,otherwise sensitivityanalysis based on thediversity of judgmentscan be performed.

This method can allowstakeholderengagement.

conducted. Thismethod can be timeand resource intensiveif a semi-structuredinterview process isused. If time is aconstraint, a writtensurvey can be used.

Literature review Based on a literaturereview, the modelers'or experts'interpretation ofstorylines is used toinfer model inputs andassumptions. Theinference is usuallydone by adapting thevariables for currentprojections to thedeveloped storylines

Depends on the wayvariables are reported.Uncertainty can beinherent if studies areprobabilistic,otherwise sensitivityanalysis can beperformed.

Not generallyapplicable.

Usually not resourceand time intensive.However, consensusbetween narrativedevelopers andquantifiers can behard to reach and aniterative process ofcommunicationbetween modelers orexperts andstakeholders canmake the processtime-intensive.

Provides direct inputto a simulation model.

(Promper et al., 2014;Sohl et al., 2014)

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2010

2.1.2. Advantages and disadvantagesApplying SD methodology to scenario narratives can shed light

on inconsistencies, provide a numerical representation of sce-narios, and allow the consideration of unexpected events that aredifficult to model otherwise (Schmitt Olabisi et al., 2010). Itsapplication can be extended to both narrative construction andscenario quantification. In terms of scenario purpose, it findsgreatest use in the evaluation of top-down policy choices andprescriptive actions. The SD method can also: (i) integrate multiplestakeholders, experts, disciplines, processes, and policy pre-scriptions, and (ii) identify leverage points for policy making (Kellyet al., 2013). Stakeholders and experts can easily be engaged in acollaborative workshop setting to use this method. Further, withavailable software, modelers can create a user-friendly interfacethat enables stakeholders to run their own scenarios. For example,users can typically adjust model parameters through drop-downmenus or “sliders.”

The main disadvantage of the SD method is the frequentinability to ensure the numerical accuracy of outputs, eitherbecause of the potential for faulty assumptions or the lack ofavailable data for validation. Therefore, using SD as a translationmethod is most suitable when the purpose is to understand thepossibility of dynamic changes and unintended consequences,rather than to predict precise model outputs for specific systemvariables.

2.2. Fuzzy cognitive maps

2.2.1. Use in translationFuzzy Cognitive Maps (FCMs) were first developed by Kosko

(1986) as a means to map and quantify fuzzy relationships be-tween system variables. Although originally introduced as a tool foranalyzing uncertainties in relationships and the resulting effects onsystem behavior, many recent applications have emphasized theuse of FCMs as a systems-thinking approach. In this capacity, it isparticularly applicable to scenarios that are characterized by mul-tiple co-evolving drivers in which the objective is to understandsystem resilience to change (Diniz et al., 2015; Gray et al., 2015; Kok,2009). Kosko (1986) explains that FCMs are especially appropriatefor exploring causality when the knowledge domain is soft; that is,

when both the variables of the system and their relationships arefundamentally fuzzy. As a means for representing qualitative be-liefs about the current or future state of a system, FCMs are relevantto land use scenario planning and analysis (Jetter and Kok, 2014;Kok, 2009). They can be applied to problems at differentgeographical scales, from local to global (Jetter and Kok, 2014), andthus can be useful for exploring a variety of policy measures andtheir implications for the social-ecological resilience of systems.This makes them a potentially valuable tool for policy makers.

In their role as a translation tool between qualitative scenariosand quantitative models, FCM construction begins by identifyingthe drivers of change described in the narratives and mapping thedrivers and relationships between them. Just as in the system dy-namics approach, diagrams are constructed for this purpose. Thetranslation can be done by experts alone or in combination withstakeholders; however the biggest advantages result from stake-holders concurrently developing narratives and their FCM repre-sentations (van Vliet et al., 2010). Structured, quantitative surveyscan help establish parameters defining the strength of relationshipsbetween drivers (van Vliet et al., 2010). Respondents start withsigns to indicate causality (þor e) and then assess the strength ofthe causality starting with a subjective scale that is subsequentlytranslated into 0 to 1 wt. (Fig. 3).

2.2.2. Advantages and disadvantagesSimilar to the SD methodology, the main advantages of FCM are

its ability to: (i) lucidly represent a complex system and provideinsight on feedbacks; (ii) provide scenario enrichment throughconcurrent development of narratives and their FCMs with stake-holders; and (iii) explicitly represent drivers and model assump-tions (Kok, 2009). It can also be explained to stakeholders easily,making it particularly useful when time and resources are aconstraint, and has the ability to allow users to be creative throughhigh level integration while still requiring concrete semi-quantitative information (Kok, 2009). Like the SD methodologyfor scenario purpose, it finds greatest use in evaluating top-downpolicy choices and prescriptive actions.

The largest disadvantage of FCM is that relationships are semi-quantified, thus also providing only a semi-quantitative outputfrom simulating an FCM. Another disadvantage is that the output

Table 2Decision matrix for choosing a translation method. Subjective rankings (none, low, med, high) correspond to degree of fulfillment of each method with each criterion, with‘high’ being the most and ‘low’ being the least.

Fig. 2. A causal loop diagram from the Minnesota 2050 project where the main focuswas on the interactions between energy demand, fuel type, and land use. Positive (þ)arrows represent a cause-and-effect relationship in which the two variables change inthe same direction. The existence of a positive feedback loop on the left side of thediagram indicates that changes in any of these variables are reinforcing. Reproducedfrom Schmitt Olabisi et al., 2010.

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 11

shows values of drivers as a function of number of iterations andnot as a function of time, though, according to Kok (2009), thisdrawback can be partly overcome by including processes that act at

approximately the same time scale. Further, significant data pro-cessing may be required to un-normalize the relationships to makethe data amenable to use in other simulation models. Lastly, alimitation of the method may be its need to be performed in aworkshop setting. FCMs lose their salience when constructedindependently.

2.3. Agent-based modeling

2.3.1. Use in translationIn the LULC literature, agent-basedmodeling (ABM) is defined as

a discrete-event simulation composed of heterogeneous andautonomous decision making entities (called ‘agents’) that interactwith each other and act upon their shared land use environment(Ligmann-Zielinska and Jankowski, 2007; Parker et al., 2003).Agents process information derived from a changing environmentand use it to make decisions about their actions (Janssen, 2005;Parker et al., 2003). These actions have an effect on land use andon the behavior of other agents in the system. Constructing an ABMfor LULC first requires a qualitative content analysis step that in-cludes the identification of agents, their attributes and actions, andthe environment. This is followed by what we refer to as the

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2012

“translation” step: the construction of a causal diagram that depictsthe major agent interactions and system processes leading to land-use change (Fig. 4). This translation can be performed with stake-holders, who may represent the different actors in the system, andcan benefit from the use of the ARDI (Actors-Resources-Dynamics-Interactions) technique (Etienne et al., 2011). The translation canalso be performed in combination with one or more of the othertranslation methods highlighted here (e.g., role playing games,pairwise comparison, and/or questionnaire surveys and in-terviews). These methods lead to the characterization of behavioralrules as well as the necessary inputs to the simulation model.

2.3.2. Advantages and disadvantagesThe major advantage of the ABM method is its ability to more

faithfully represent the decision processes of many agents and thenassess the effects of their micro-scale actions and interactions onmacro-scale phenomena (Parker et al., 2008). It is thereforeparticularly suitable for evaluating lower-level actions and theiranticipated outcomes. While ABMs can also be used for evaluatingtop-down policy choices and prescriptive actions, other methodslike SDs and FCM might be more easily used for this purpose, thusavoiding the need to embrace unnecessary complexity. The use ofother complementary methods can allow construction of narrativescenarios and model quantification. ABM is often used in com-panion modeling approaches to serve as a social learning tool,helping stakeholders to understand the impacts of their actions onthe future of the land (Barreteau et al., 2003; Bousquet et al., 1999).Further, the engagement of local stakeholders in the constructionand use of the simulation tool can help facilitate the emergence ofcollective plans and actions, and thus the method finds great use ina collaborative workshop setting with experts and stakeholders.

The major criticism of the ABM methodology lies in thecomplexity associated with modeling multiple decision makingagents within the context of a complex human-natural environ-ment (Parker et al., 2002). This limits the use of method to evalu-ating anticipated outcomes associated with lower-level actions.Detailed quantitative and qualitative information is often requiredto parameterize the model, and such information may be lacking atthe relevant spatial scales (Kelly et al., 2013; Le Page et al., 2014).This lack of data can also create a problem for model validation.

Fig. 3. An FCM developed to describe the system of drivers of deforestation for th

Further, the methodology by itself finds best use when performedin a collaborative setting with stakeholders and experts. This is anecessity as the methodology may be hard for stakeholders tounderstand. While the ABM methodology can be used indepen-dently, this poses a great challenge especially when it comes tocharacterization of behavioral rules and determination of othermodel inputs.

2.4. Fuzzy sets

2.4.1. Use in translationThe fuzzy sets translation method has its roots in fuzzy set

theory, which was first proposed by Zadeh (1965, 1965), and laterGoguen (1969), proposed to study fuzziness contained in humanlanguage by using a mathematical construct that extended thegeneral notion of a set. In theory, such a construct lends itself verywell to translating narratives to simulation model inputs that arequantitative in nature. Fuzzy set theory has found wide applicationin many areas of environmental science (Cornelissen et al., 2001;Geyer-Schulz, 1995) and was used in translation for land useassessment by Alcamo et al. (2008) and Onigkeit et al. (2007) aspart of the Glowa Jordan River scenario exercise. Based on theseexperiences, further applications included work in multiple casestudies in Europe (Kok et al., 2015). Like FCMs, the fuzzy setsmethod provides a mathematical approach for studying vagueconceptual phenomena (Zimmermann, 2010). Significant work hasalso been done on mapping and geovisualization of fuzzy sets(Ahlqvist, 2004; Lucieer and Kraak, 2004). Fuzzy sets are alsoapplicable to problems that span across various scales. However,they have typically been applied to problems at a global scale whenthere are a large number of interacting drivers of change, thusmaking their translation through a systems thinking approachcomplicated and time consuming.

According to Alcamo (2009) there are five steps involved intranslating storylines to model inputs using fuzzy sets: (i) makesubjective statements about rates of change of all importantdrivers; (ii) derive a translation key that allows subjective state-ments to be translated into numbers; (iii) combine the obtained setof numbers into a membership function; (iv) use membershipfunctions directly by linking with probability density functions or

e Brazilian Amazon as developed by stakeholders (Adapted from Kok, 2009.).

Fig. 4. An overview of an ABM-based translation framework, assuming that the focus of the model is on land use composition and configuration, producing output LULC maps atdifferent time steps of model execution.

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 13

defuzzifying the output to obtain inputs for simulationmodels; and(v) use the outputs of simulation models to enrich storylines.

2.4.2. Advantages and disadvantagesThe most important advantage of this method is that stake-

holder knowledge can be used for direct model parameterization,without the need for interpretation. In terms of scenario purpose,fuzzy sets are best used when the overall goal is to evaluate top-down policy choices and prescriptive actions. Other advantagesinclude the ease with which fuzzy sets can be integrated into thescenario development process and its usefulness when performedin a collaborative setting. Stakeholders using fuzzy sets can easily

be involved not just in the process of translation but also in modelparameterization, and their input is especially valuable in situa-tions in which drivers of change are directly influenced by stake-holder choices. In many situations, subjective scales along withmembership functions are used to facilitate translation. Themembership functions allow scenario developers an easy way toconvert between quantitative model variables and linguisticstatements in a reproducible manner.

Themajor disadvantage of the fuzzy set method is that it focusessolely on quantifiable parameters. This leads to its inability toincorporate feedbacks and structural relationships in the narrative.It finds almost no use in narrative construction, and relies on its use

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2014

in a collaborative workshop atmosphere, thus limiting its use in anindependent setting. Also, a test for scenario consistency can onlybe performed after running the associated simulation model.Further, membership functions do not fully quantify all aspects ofthe narrative, and fail to take into account “softer” drivers of change(e.g., cultural identity).

2.5. Pairwise comparison

2.5.1. Use in translationThe pairwise comparison method was first used in the evalua-

tion of scenarios by Saaty and Rogers (1976). Their approach allowsverbal judgments of experts to be translated into a numerical scale.Further development led to the addition of the analytic hierarchyprocess which allows ranking of judgments of relative importance(Saaty, 1980). More recently, the pairwise comparison approach hasfound applicability in agricultural land-use scenario analysis(Abildtrup et al., 2006). This methodology allows translation ofqualitative beliefs into quantitative inputs for simulation models byfollowing a simple relative ranking procedure and eigenvectorcalculation (Rosato and Giupponi, 2003). Like fuzzy sets, theapplicability of pairwise comparison stretches across variousgeographical scales but is typically more useful for global orregional problems in which the set of objectives are broader thanthose justifying a SD approach.

The Pairwise Comparison process follows four steps: (i) sub-jective pairwise comparison for each driving force or parameteracross all scenarios; (ii) translation of subjective measures intonumerical values using the scales proposed by Saaty (1980); (iii)evaluation of the consistency of translation; and, if consistent, (iv)derivation of eigenvectors for each driver (Rosato and Giupponi,2003). An appropriate scale must be pre-determined for eachdriver, and is usually based on the judgment of scenario builders. Ageneral rule that can be followed is that the choice of scale shoulddepend on the expected effect of a particular driver on the outcomeof the scenario: the smaller the anticipated effect, the smaller thescale range (Abildtrup et al., 2006). Historically, this method hasbeen employed only with experts and is considered an expert-judgment technique (Abildtrup et al., 2006). However, stake-holders have been involved in translation, especially in the phase ofranking drivers across scenarios (Strager and Rosenberger, 2006).The numerical evaluations and consistency checks for judgmentscan be performed with an available software tool. Abildtrup et al.(2006) provide a detailed example of the implementation of themethod.

2.5.2. Advantages and disadvantagesThe major advantage of this method is that it can be easily un-

derstood by experts coming from diverse backgrounds, as themathematical construct is fairly simple to grasp (Saaty, 1980). Likefuzzy sets, the method finds greatest use when used when thescenario purpose is to evaluate top down policy choices and pre-scriptive actions. Further, it explicitly involves a consistency checkto evaluate experts’ judgments (Abildtrup et al., 2006). An incon-sistency could hint at vagueness in narrative descriptions and theneed to enrich certain aspects of scenarios. The outputs from themodel can directly inform simulation models. The method is reallyeasy to understand, and can be performed directly withstakeholders.

The disadvantage of the method is that it does not offer a robustway to check consistency across narrative scenarios. It offers littlehelp in the construction of narrative scenarios and relies on its usein a collaborative setting. Further, the scale must be pre-determined and is based completely on the judgment of the sce-nario developers (Abildtrup et al., 2006). The process can also be

time-consuming, especially when working through various com-binations in trying to achieve consistency across experts (Abildtrupet al., 2006).

2.6. Bayesian statistical reasoning

2.6.1. Use in translationBayesian statistical reasoning as applied to scenario translation

(Kemp-Benedict, 2010) has its foundations in Bayes' rule of condi-tional probability. According to Bayes’ rule, the posterior distribu-tion of a particular model parameter z, given a particular scenario S,is written as:

PðzjSÞfPðSjzÞPðzÞ (1)

where P(z) is a prior probability specifying the distribution for theparameter in a reference case, and P(Sjz) is a conditional proba-bility, which can be thought of as a measure of the degree of con-fidence that scenario S is occurring given information on the valueof z.

The translation process thus begins by specifying prior distri-butions for each of the parameters in the reference case, which canbe a baseline or a business-as-usual scenario (Kemp-Benedict,2010). Historical data are generally used as the basis for the refer-ence case, and stakeholder or expert elicitations can be used toconstruct prior distributions that reflect the range of beliefs (Kemp-Benedict, 2010). The next step is to evaluate the conditional prob-ability P(Sjz) based on the knowledge and experience of stake-holders and experts. The posterior distribution of each parameterunder each scenario can then be evaluated from eq. (1), often withthe aid of available software (Kemp-Benedict, 2010).

2.6.2. Advantages and disadvantagesThe major advantage of this method is the ease with which it

can be implemented in aworkshop setting with experts. A softwareprogram can be used to quickly calculate posterior parameter dis-tributions. Depending on the simulation model used, these distri-butions can then be used directly to represent uncertainty. If thesimulation model is not compatible with stochastic inputs, thenposterior means might be used as inputs. In terms of the scenariopurpose, it's best used for evaluating top down policy choices andprescriptive actions.

The major disadvantage of this method is in its limited appli-cability with stakeholders. Some understanding of the ideas ofprobability and likelihood distributions is necessary, though ex-perts may be able to design the translation process in a way thatfacilitates stakeholder engagement. The method also has limitedapplicability in understanding feedbacks or testing narrative con-sistency, and is constrained in its ability to quantify or represent thesofter drivers of change in the narrative. It offers limited extensi-bility in terms of narrative construction.

2.7. Bayesian network

2.7.1. Use in translationThe idea of constructing graphical influence diagrams and then

populating them with conditional probability tables to developBayesian networks (BN) was first formalized by Pearl (1985, 1985)constructed a BN with the goal of devising a computationalmodel for making inferences about human reasoning. In the field ofland use change, participatory modeling approaches with BNs haveexplored the role of stakeholders as information providers in theabsence of empirical data to define conditional probability tables,as well as in the construction of the influence diagrams themselves.

A BN consists of two key components: (i) an influence diagram

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 15

that describes underlying relationships in a system and (ii) a set ofconditional probability tables, which define the quantitative re-lationships between each variable and its parents. Together, thesecomponents decompose a complex causal chain into a series oflocal relationships. These relationships can be characterized usingvariety of sources including stakeholder beliefs, experts, and liter-ature review (Barton et al., 2012; Marcot et al., 2001). The finalnodes describing the land use outcome are generated by propa-gating uncertainty through the network using probability calculusor available simulation software (Aalders, 2008; Meyer et al., 2014).These outputs can then feed into more detailed simulation models(e.g., cellular models of land use change). Since the outcome of theBN is the land-use change phenomena researchers are interestedin, the translation step can be defined as the construction of theinfluence diagram and specification of conditional probabilitiesthat drive land use change.

In one recent application of BNs to land use change, researchersin Maine, USA, developed a framework for using BNs with a cellularautomata model to translate stakeholder-derived qualitative sce-narios of land use change into quantitative, spatially explicit futureland use simulations (Fig. 5) (Meyer et al., 2014). The BNs, whichprovided pixel-level suitability assessments (Fig. 6), were used asinput into the cellular automata model, which applied pixel andpatch-level transitions based on stakeholder-derived future sce-narios. Using results from this study, land managers and policymakers were able to: (i) identify high-value conservation, economicdevelopment, and natural resource assets; (ii) evaluate potentialcurrent and future conflicts and opportunities between land uses;and (iii) envision the impacts of alternative policy decisions onfuture land use, both at the local- and landscape-levels.

2.7.2. Advantages and disadvantagesThe major benefits of the BN approach are the ability to rapidly

update model outputs as new data become available and the abilityto re-calibrate the model to a different set of stakeholder inputs(e.g., translation and application to another study area). In terms ofscenario purpose, it is valuable in evaluating top-down policychoices and prescriptive actions; but can also offer some insightinto the anticipated outcomes of certain lower-level actions. Otheradvantages lie in its ability to combine different sources ofknowledge, including expert as well as stakeholder beliefs (Borsuket al., 2004; Marcot et al., 2006; Uusitalo, 2007). This also makesBNs suitable for applications where data are incomplete. This isespecially valuable for land use change applications, as land usedata often contain missing values. The outputs from influence di-agrams and conditional probability tables can inform not just BNbut also cellular models where transitions (e.g., in land use change)are based on probabilities. BNs can thus be used for developingquantitative assessments for model-based inputs.

The disadvantages of BN lie in the process of translation withstakeholders and experts. It may be difficult to obtain informationin the form of conditional probabilities, especially if stakeholdershave difficulty conceptualizing the idea of conditional distributions(Morgan et al., 1992). This can be overcome if the translation pro-cess is carefully designed to solicit relevant stakeholder knowledge.Further, BN are most often used to study specific systems or sub-systems and a specific change associated with the system. If nar-ratives describe multiple land use changes, BN modeling of all thechanges can be a tedious task. Further, BNs fail to account forfeedbacks in the system, being acyclic in structure (Jensen, 2001).Several networks can be constructed to model temporal and spatialdynamics of feedback; however, this task may prove to be tedious(Jensen, 2001). While BNs can be constructed independently, it isdifficult especially when it comes to assessing conditional proba-bility distributions.

2.8. Role playing games

2.8.1. Use in translationRole Playing Games (RPGs) can be defined as “the performance

of an imaginary or realistic situation played by people with givenroles in order to analyze behavioral patterns” (Shaftel and Shaftel,1967). RPGs can be used to create narratives, identify drivers ofchange, and parameterize simulation models for coupled humanand natural systems (Castella et al., 2005; Pak and Brieva, 2010;Washington-Ottombre et al., 2010). Role playing games wereused in land-use management and planning by Commere (1989).Since then, they have gained popularity in the field of land-usechange and have been employed as a social learning and collec-tive management tool (Castella et al., 2005; Dumrongrojwatthanaet al., 2011; Souch�ere et al., 2010). Stand-alone RPGs have beenused to investigate the results of stakeholder actions on land-usechange by generating maps of the future, and in mediating nego-tiations in ecological systems that require resource sharing (Dunget al., 2009; Pak and Brieva, 2010). More recently, role playinggames have been used to parameterize agent based models byidentifying specific drivers of change (Castella et al., 2005; Le Pageet al., 2014; Washington-Ottombre et al., 2010). Lamarque et al.(2013) used role playing games to create quantitative maps ofland use change from qualitative storylines and descriptions offarmers’ decision making.

The use of RPG in translation is most compatible with agent-based and multi-agent simulation models, given the ability ofboth to model individual behavior (Barreteau et al., 2001; Bousquetet al., 1999). The translation process begins with the co-creation of agame with heterogeneous stakeholders that represents the base-line scenario to be explored (D'Aquino et al., 2002). This is followedby identification of alternate scenarios using a modified version ofthe RPG or a computer ABM. It is important to identify relevantstakeholders to create a game that appropriately simulates a real-world representation of their roles and responsibilities. Whenstakeholders are used to construct alternate scenarios, the basicgame needs to be modified to accommodate their proposedchanges. A game consistent across all scenarios can be generatedwith variable constraints and decision rules based on the scenarios.Stakeholder actions during the game can be interpreted as inputs tosimulation models or can serve to create stand-alone quantitativemaps of land-use change (D'Aquino et al., 2002; D'Aquino et al.,2003). Usually, a series of successive versions of the game areused with stakeholders not just to simulate various proposed sce-narios but also to facilitate social learning, consensus building, andobjective exploration. While the type of game created variesdepending on the system being studied, some standard role playinggames exist in the literature and have been adapted for variousapplications in land use change (Campbell and Palutikof, 1978;Meadows et al., 1989).

2.8.2. Advantages and disadvantagesThe biggest advantage of RPGs is the ability to accommodate the

range of behaviors, decision-making, and adaptation through directinvolvement of stakeholders (Lamarque et al., 2013). Narrativedescriptions are easily communicated through the medium of agame to stakeholders and the uncertainty inherent in narrativedescriptions is easily accommodated (Lamarque et al., 2013; Martinet al., 2011). It is therefore useful to be used as a narrative con-struction tool. RPGs offer an easy way to understand the individualbehavioral drivers that contribute to land management change(Naivinit et al., 2010; Pak and Brieva, 2010). It is an effective tool forsocial learning by fostering the exchange of knowledge betweenland users and scientists (Barnaud et al., 2010;Dumrongrojwatthana et al., 2011; Martin et al., 2011). Model

Fig. 5. Schematic of an integrated Bayesian network/cellular automata land use scenario model (Meyer et al., 2014).

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2016

validation can be performed in several ways such as statisticalanalyses, interviews with stakeholders, and by consulting expertson proposed adaptations (Castella et al., 2005; Lamarque et al.,2013; Washington-Ottombre et al., 2010). In terms of scenariopurpose, it is especially useful when used to evaluate lower-levelpolicy choices and their anticipated outcomes, given the focus ofthe method on the human behavior and its outcome.

Despite these advantages, the method can be extremelydemanding in terms of resources and time commitment. The suc-cess of the method hinges on the degree of engagement of stake-holders and the quality of knowledge and information exchange, aswell as the compatibility of the method with simulation tools.Engaging stakeholders can be a challenging task if the rules of thegame are difficult to understand. Further, it is important that RPGsadequately represent aspects of a scenario, stimulate user crea-tivity, and provide information on behavioral drivers of changewhile also ensuring that the RPGs represent stakeholders' views ofreality, and not their views on making an interesting game. Severalruns may be necessary to achieve results that are consistent andrepeatable.

2.9. Other methods

There are other, less-formal methods of translating scenarionarratives into quantitative models, including questionnaire sur-veys, semi-structured interviews, and literature reviews. Surveysand semi-structured interviews are surveys of subjective beliefs ofexperts or stakeholders, and can allow limited stakeholderengagement while providing direct input for the simulation model

(McDaniels et al., 2012). These can also be conducted in a structuredway using a Delphi method within the context of a workshop(Dubrovsky et al., 2011). Literature review is probably the mostcommon additional approach and is often conducted by themodeler using his or her interpretation of the narratives. Thismethod limits engagement with stakeholders and tends to beheavily influenced by existing data and model studies (Dubrovskyet al., 2011).

3. Discussion

Moving from narrative scenarios to a quantitative model-basedassessment of land use change requires an approach that faithfullycharacterizes the original narratives and is reproducible andinternally consistent. For this reason, LULC modelers may want tofollow an established approach, rather than relying on their ownindependent, and often ad-hoc, judgment. With this in mind, weaim to facilitate the choice of approach by revisiting the methodsreviewed above according to four practical considerations(Table 2): (i) purpose of the scenario process, (ii) potential to extendthe approach to include narrative creation and/or quantitativemodeling, (iii) type of participation desired, and (iii) type ofcompatible simulation model. We describe each of these consid-erations and our basis for scoring the approaches below.

3.1. Scenario purpose

The choice of a model and/or translation method should bestrongly influenced by the level of analysis necessary in the

Fig. 6. One of the Bayesian network diagrams of Meyer et al. (2014) used to assess landscape conservation suitability. In this example, specific findings have been entered for eachfactor to show the overall probability of suitability based on those inputs (Meyer et al., 2014).

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 17

scenario development process. Key developments in both the bio-physical and socio-economic aspects of the considered scenariosmay occur at either state/regional/national level or at the field/community level (Castella et al., 2007). Similarly, policy questionsbeing asked of the analysis may pertain to either top-down policyactions and prescriptions, or lower-level actions and anticipatedresponses to policy. The ‘top-down’ emphasis is appropriate whenland use changes are largely determined by exogenous forces, suchas changes in global demand and market conditions (Verburg andOvermars, 2009). A ‘bottom-up’ or lower-level perspective isneeded when changes are the result of local land ownership andthe corresponding management decisions being made. Hybridcases may also exist, where lower level pilot experiments are scaledup to change the top down policy making. As indicated in Table 2,the translation methods reviewed are largely oriented toward oneor the other of these perspectives, with none being ideal for both,and therefore we recommend a mixed methods approach forhybrid cases.

3.2. Extensibility

Many of the scenario translation methods can also be useful tothe stakeholder-driven process of constructing the narrative sce-nario and/or a model-based process of scenario quantification.Explicitly graphical methods, such as SD and FCM, can jointlyengage experts and stakeholders in the process of envisioning anddefining narrative scenarios (Mavrommati et al., 2014; SchmittOlabisi et al., 2010; Yu et al., 2011). Methods emphasizing individ-ual decision-making, such as agent-based modeling and role

playing games, can help in identifying larger scale emergent pat-terns needed to distinguish diverse scenarios (Gerst et al., 2013).

Process-based translation methods, including system dynamicsand agent-based modeling, can be fully parameterized to generatesimulations that quantify the scenarios of interest. Similarly,graphical methods of reasoning, such as fuzzy cognitive maps andBayesian networks, can be specified numerically using empirical orsubjective reasoning under uncertainty. These then yield scenarioquantifications that account for lack of knowledge and/or inherentrisk.

The more nonrepresentational methods, including fuzzy sets,pairwise comparison, and Bayesian reasoning, are useful primarilyfor translating between qualitative narratives and quantitativemodels developed by other means. Similarly, surveys, interviews,and literature review are not particularly extensible beyond thepurposes of translation (Table 2).

3.3. Type of participation

Too often, the task of translating narrative scenarios to quanti-tative model formulations and inputs has been left to technicalmodelers. Yet, as indicated in Table 2, most of the translationmethods reviewed here are appropriately applied when expertsand stakeholders work cooperatively. The participation of a broadbase of experts across particular areas of expertise or disciplinesexpands the range of methods that could be used. Likewise,stakeholder involvement is essential for successful application ofsome translation methods, including role-playing games, surveys,and interviews. Other methods, including system dynamics, agent-

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2018

based modeling, fuzzy cognitive maps, fuzzy sets, and Bayesiannetworks, can accommodate stakeholder participation, and espe-cially benefit from experts and stakeholders working together.Because of their complexity, methods that require probabilisticreasoning need to engage stakeholders in innovative ways, such asparticipatory modeling with Bayesian networks (Meyer et al.,2014). Still other methods, like literature reviews offer no oppor-tunity for stakeholder engagement.

Scenario translation can serve as an opportunity for researchersand stakeholders to exchange points of view. Yet, the time, effort,conflict resolution, and communication skills demanded of aparticipatory approach to scenario translation may make thetransaction costs too much to bear (Sayer et al., 2013). Somestakeholders may also have difficulty understanding scenarios andmodels (Becu et al., 2008) and, therefore, the translation process.Field studies have revealed the importance of stakeholder under-standing of the method for them to trust it and use it beyond lipservice. Modelers who plan to engage stakeholders in scenariotranslation ought to take the time to learn from existing bestpractices in participatory modeling, including flexibility, temporaland spatial openness, and transparency (Voinov and Bousquet,2010).

3.4. Compatible model types

As described in the sub-section on extensibility above, sometranslation methods can either be used directly as quantitativemodels (e.g., system dynamics, agent-based, fuzzy cognitive maps,and Bayesian networks) or are naturally paired with certain modelmethodologies (e.g., fuzzy cognitive maps and system dynamics,Bayesian reasoning and statistical models, or role-playing gamesand agent-based models). Others, including fuzzy sets, pairwisecomparison, surveys and interviews, and literature review can beused with most, if not all, types of simulation models.

4. Conclusions

Our review of methods for translating between narrative andquantitative scenarios, including the guidance given in Table 2,should allow those involved in the scenario development process tochoose a method appropriate to their scale, purpose, level ofparticipation, and project resources. For example, one can see fromTable 2 that if the goal of a project is to explore the outcomes oflower-level land-use decisions by developing both qualitative andquantitative scenarios in a combined expert/stakeholder workshopsetting, then agent-based modeling may be the best choice. If, onthe other hand, the goal is to explore higher-level, top-down pol-icies by developing a quantitative model informed primarily byexperts, then system dynamics, fuzzy cognitive maps, or Bayesiannetworks would be most appropriate. A mixed methods approachis most appropriate in situations of a hybrid case, where lower levelpilot experiments are scaled up to inform the top down policymaking. The table also reveals that relying on the existing literaturealone, often the default approach for many modelers, may not bethe optimal scenario translation choice given the study objectives.

Practically, in some situations it may be best to combine mul-tiple options in amixedmethods approach. In the PRELUDE project,for example, stakeholders developed influence diagrams of themain drivers in the narratives and determined a subjectivemagnitude of change for each of the drivers (Volkery et al., 2008).These qualitative values were subsequently translated into a nu-merical value based on existing scenarios and past data. Hence,literature reviews and semi-structured interviews were used incombination with Bayesian networks and a variation of fuzzycognitive maps to quantify narratives. We expect such creative

exploitation of the individual advantages of many of the methodsreviewed here to be a practical approach for many real-world as-sessments of land use change going forward.

Acknowledgements

This researchwas supported in part by the Scenarios Society andSolutions Research Coordination Network (Grant No. NSF-DEB-13-38809). We thank Kathy Fallon Lambert and Katie Theoharides fortheir efforts leading the S3 RCN and organizing the Scenarios toSimulation workshop in Hanover, NH in February 2015. Weappreciate the expertise and involvement of the workshop partic-ipants. Additional support was provided by the National ScienceFoundation through the New Hampshire Experimental Program toStimulate Cooperative Research (NH EPSCoR) Ecosystems and So-ciety Project (Grant # EPS 1101245). We also thank Cameron Wake,Alexandra Thorn, Curt Grim, Barbara Wauchope and Clay Mitchellfor providing the scenarios from the Ecosystems and Society Projectfor use in the workshop.

References

Aalders, I., 2008. Modeling land-use decision behavior with Bayesian belief net-works. Ecol. Soc. 13 (1), 16.

Abildtrup, J., Audsley, E., Fekete-Farkas, M., Giupponi, C., Gylling, M., Rosato, P.,Rounsevell, M., 2006. Socio-economic scenario development for the assessmentof climate change impacts on agricultural land use: a pairwise comparisonapproach. Environ. Sci. Policy 9 (2), 101e115.

Ahlqvist, O., 2004. A parameterized representation of uncertain conceptual spaces.Trans. GIS 8 (4), 493e514.

Alcamo, J., 2009. The SAS approach: combining qualitative and quantitativeknowledge in environmental scenarios. In: Joseph, A. (Ed.), Environmetal Fu-tures. The Practice of Environmental Scenario Analysis. Elsevier, pp. 123e150.

Alcamo, J., Acosta-Michlik, L., Carius, A., Eierdanz, F., Klein, R., Kr€omker, D.,T€anzler, D., 2008. A new approach to quantifying and comparing vulnerabilityto drought. Reg. Environ. Change 8 (4), 137e149.

Barnaud, C., Van Paassen, A., Trebuil, G., Promburom, T., Bousquet, F., 2010. Dealingwith power games in a companion modelling process: lessons from communitywater management in Thailand highlands. J. Agric. Educ. Ext. 16 (1), 55e74.

Barreteau, O., Antona, M., d'Aquino, P., Aubert, S., Boissau, S., Bousquet, F., Dar�e, W.,Etienne, M., Le Page, C., Mathevet, R., 2003. Our companion modellingapproach. J. Artif. Soc. Soc. Simul. 6 (2), 1.

Barreteau, O., Bousquet, F., Attonaty, J.-M., 2001. Role-playing games for opening theblack box of multi-agent systems: method and lessons of its application toSenegal river Valley irrigated systems. J. Artif. Soc. Soc. Simul. 4 (2), 5.

Barton, D.N., Kuikka, S., Varis, O., Uusitalo, L., Henriksen, H.J., Borsuk, M., de laHera, A., Farmani, R., Johnson, S., Linnell, J.D., 2012. Bayesian networks inenvironmental and resource management. Integr. Environ. Assess. Manag. 8 (3),418e429.

Becu, N., Neef, A., Schreinemachers, P., Sangkapitux, C., 2008. Participatory com-puter simulation to support collective decision-making: potential and limits ofstakeholder involvement. Land Use Policy 25 (4), 498e509.

Borsuk, M.E., Stow, C.A., Reckhow, K.H., 2004. A Bayesian network of eutrophicationmodels for synthesis, prediction, and uncertainty analysis. Ecol. Model. 173 (2),219e239.

Bousquet, F., Barreteau, O., Le Page, C., Mullon, C., Weber, J., 1999. An environmentalmodelling approach: the use of multi-agent simulations. In: Blasco, F., Weil, A.(Eds.), Advances in Environmental and Ecological Modelling. Elsevier, Paris,pp. 113e122.

Campbell, D.J., Palutikof, J., 1978. Allocation of Land Resources in Semi-arid Areas: aSimulation Based on the East African Experience. Discussion Paper 262. Insti-tute for Development Studies, University of Nairobi, Nairobi.

Carpenter, S.R., Booth, E.G., Gillon, S., Kucharik, C.J., Loheide, S., Mase, A.S.,Motew, M., Qiu, J., Rissman, A.R., Seifert, J., 2015. Plausible futures of a social-ecological system: Yahara watershed, Wisconsin. USA. Ecol. Soc. 20 (2), 10.

Cash, D., Clark, W.C., Alcock, F., Dickson, N.M., Eckley, N., J€ager, J., 2002. Salience,Credibility, Legitimacy and Boundaries: Linking Research, Assessment and De-cision Making. Harvard University.

Castella, J.-C., Kam, S.P., Quang, D.D., Verburg, P.H., Hoanh, C.T., 2007. Combiningtop-down and bottom-up modelling approaches of land use/cover change tosupport public policies: application to sustainable management of natural re-sources in northern Vietnam. Land Use Policy 24 (3), 531e545.

Castella, J.-C., Trung, T.N., Boissau, S., 2005. Participatory simulation of land-usechanges in the northern mountains of Vietnam: the combined use of anagent-based model, a role-playing game, and a geographic information system.Ecol. Soc. 10 (1), 27.

Commere, R., 1989. Le d�eveloppement local en milieu rural. Universit�e de Saint,p. 221.

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e20 19

Cornelissen, A., Van den Berg, J., Koops, W., Grossman, M., Udo, H., 2001. Assess-ment of the contribution of sustainability indicators to sustainable develop-ment: a novel approach using fuzzy set theory. Agric. Ecosyst. Environ. 86 (2),173e185.

D'Aquino, P., Barreteau, O., Etienne, M., Boissau, S., Aubert, S., Bousquet, F., LePage, C., Dar�e, W., 2002. The Role Playing Games in an ABM participatorymodeling process: outcomes from five different experiments carried out in thelast five years, Integrated assessment and decision support. In: 1st BiennialMeeting of the International Environmental Modelling and Software Society.iEMSs, Lugano, Switzerland, pp. 24e27.

D'Aquino, P., Le Page, C., Bousquet, F., Bah, A., 2003. Using self-designed role-playinggames and a multi-agent system to empower a local decision-making processfor land use management: the SelfCormas experiment in Senegal. J. Artif. Soc.Soc. Simul. 6 (3).

Diniz, F.H., Kok, K., Hoogstra-Klein, M.A., Arts, B., 2015. Mapping future changes inlivelihood security and environmental sustainability based on perceptions ofsmall farmers in the Brazilian Amazon. Ecol. Soc. 20 (2), 26.

Dubrovsky, M., Simon, K., Stuch, B., Kok, K., Harrison, P., 2011. Report on the Eu-ropean Driving Force Database for Use in the Integrated Assessment Platform,the CLIMSAVE Project Climate Change Integrated Assessment Methodology forCross-sectoral Adaptation and Vulnerability in Europe.

Dumrongrojwatthana, P., Le Page, C., Gajaseni, N., Tr�ebuil, G., 2011. Co-constructingan agent-based model to mediate land use conflict between herders and for-esters in northern Thailand. J. Land Use Sci. 6 (2e3), 101e120.

Dung, L., Hoanh, C.T., Le Page, C., Bousquet, F., Gajaseni, N., 2009. Facilitating dia-logue between aquaculture and agriculture: lessons from role-playing gameswith farmers in the Mekong Delta, Vietnam. Water Policy 11, 80e93.

Etienne, M., Du Toit, D.R., Pollard, S., 2011. ARDI: a co-construction method forparticipatory modeling in natural resources management. Ecol. Soc. 16 (1), 44.

Ford, F.A., 1999. Modeling the Environment: an Introduction to System DynamicsModels of Environmental Systems. Island Press, Washington, DC.

Gerst, M., Wang, P., Borsuk, M., 2013. Discovering plausible energy and economicfutures under global change using multidimensional scenario discovery. Envi-ron. Model. Softw. 44, 76e86.

Geyer-Schulz, A., 1995. Fuzzy Rule-based Expert Systems and Genetic MachineLearning. Physica-Verlag, Heidelberg.

Goguen, J.A., 1969. The logic of inexact concepts. Synthese 19 (3), 325e373.Gray, S.A., Gray, S., De Kok, J.L., Helfgott, A.E., O'Dwyer, B., Jordan, R., Nyaki, A.,

Analyzing, F., 2015. Using fuzzy cognitive mapping as a participatory approachto analyze change, preferred states, and perceived resilience of social-ecologicalsystems. Ecol. Soc. 20 (2), 11.

Houet, T., Aguejdad, R., Doukari, O., Battaia, G., Clarke, K., 2016. Description andValidation of a “non Path-dependent” Model for Projecting Contrasting UrbanGrowth Futures. European Journal of Geography, Cybergeo.

Janssen, M.A., 2005. Agent-based modelling. Model. Ecol. Econ. 1, 155e172.Jensen, F.V., 2001. Bayesian Networks and Decision Graphs. Statistics for Engi-

neering and Information Science, 2 ed. Springer, Verlag New York.Jetter, A.J., Kok, K., 2014. Fuzzy cognitive maps for futures studiesdA methodo-

logical assessment of concepts and methods. Futures 61, 45e57.Kelly, R.A., Jakeman, A.J., Barreteau, O., Borsuk, M.E., ElSawah, S., Hamilton, S.H.,

Henriksen, H.J., Kuikka, S., Maier, H.R., Rizzoli, A.E., 2013. Selecting among fivecommon modelling approaches for integrated environmental assessment andmanagement. Environ. Model. Softw. 47, 159e181.

Kemp-Benedict, E., 2010. Converting qualitative assessments to quantitative as-sumptions: Bayes' rule and the pundit's wager. Technol. Forecast. Soc. Change77 (1), 167e171.

Kemp-Benedict, E., 2013. Going from Narrative to Number: Indicator-driven Sce-nario Quantification, Recent Developments in Foresight Methodologies.Springer, New York, pp. 123e131.

Kok, K., 2009. The potential of fuzzy cognitive maps for semi-quantitative scenariodevelopment, with an example from Brazil. Glob. Environ. Change 19 (1),122e133.

Kok, K., B€arlund, I., Fl€orke, M., Holman, I., Gramberger, M., Sendzimir, J., Stuch, B.,Zellmer, K., 2015. European participatory scenario development: strengtheningthe link between stories and models. Clim. Change 128 (3e4), 187e200.

Kosko, B., 1986. Fuzzy cognitive maps. Int. J. Man Mach. Stud. 24 (1), 65e75.Lamarque, P., Artaux, A., Barnaud, C., Dobremez, L., Nettier, B., Lavorel, S., 2013.

Taking into account farmers' decision making to map fine-scale land manage-ment adaptation to climate and socio-economic scenarios. Landsc. Urban Plan.119, 147e157.

Le Page, C., Naivinit, W., Tr�ebuil, G., Gajaseni, N., 2014. Companion Modelling withRice Farmers to Characterise and Parameterise an Agent-based Model on theLand/water Use and Labour Migration in Northeast Thailand, Empirical Agent-based Modelling-challenges and Solutions. Springer, pp. 207e221.

Ligmann-Zielinska, A., Jankowski, P., 2007. Agent-based models as laboratories forspatially explicit planning policies. Environ. Plan. B Plan. Des. 34 (2), 316e335.

Lucieer, A., Kraak, M.-J., 2004. Interactive and visual fuzzy classification of remotelysensed imagery for exploration of uncertainty. Int. J. Geogr. Inf. Sci. 18 (5),491e512.

Marcot, B.G., Holthausen, R.S., Raphael, M.G., Rowland, M.M., Wisdom, M.J., 2001.Using Bayesian belief networks to evaluate fish and wildlife population viabilityunder land management alternatives from an environmental impact statement.For. Ecol. Manag. 153 (1), 29e42.

Marcot, B.G., Steventon, J.D., Sutherland, G.D., McCann, R.K., 2006. Guidelines fordeveloping and updating Bayesian belief networks applied to ecological

modeling and conservation. Can. J. For. Res. 36 (12), 3063e3074.Martin, G., Felten, B., Duru, M., 2011. Forage rummy: a game to support the

participatory design of adapted livestock systems. Environ. Model. Softw. 26(12), 1442e1453.

Mavrommati, G., Baustian, M., Dreelin, E., 2014. Coupling socioeconomic and lakesystems for sustainability: a conceptual analysis using lake St. Clair region as acase study. AMBIO 43 (3), 275e287.

McDaniels, T., Mills, T., Gregory, R., Ohlson, D., 2012. Using expert judgments toexplore robust alternatives for forest management under climate change. RiskAnal. 32, 2098e2112.

Meadows, D., Fiddaman, T., Shannon, D., 1989. Fish Banks. Institute for Policy andSocial Science Research. University of New Hampshire, Durham.

Meyer, S.R., Johnson, M.L., Lilieholm, R.J., Cronan, C.S., 2014. Development of astakeholder-driven spatial modeling framework for strategic landscape plan-ning using Bayesian networks across two urban-rural gradients in Maine, USA.Ecol. Model. 291, 42e57.

Millennium Ecosystem Assessment, 2003. Ecosystems and Human Well-being: aFramework for Assessment, Millenium Ecosystem Assessment: Ecosystems andHuman Well-being. World Resources Institute, Washington, p. 266.

Morgan, M.G., Henrion, M., Small, M., 1992. Uncertainty: a Guide to Dealing withUncertainty in Quantitative Risk and Policy Analysis. Cambridge UniversityPress, Cambridge, UK.

Naivinit, W., Le Page, C., Tr�ebuil, G., Gajaseni, N., 2010. Participatory agent-basedmodeling and simulation of rice production and labor migrations in North-east Thailand. Environ. Model. Softw. 25 (11), 1345e1358.

Nakicenovic, N., Swart, R., 2000. Special report on emissions scenarios. In:Nakicenovic, Nebojsa, Swart, Robert (Eds.). Cambridge University Press. 1,Cambridge, UK, p. 612.

Onigkeit, J., Lübkert, B., Alcamo, J., 2007. Quantitative Model Drivers for the GLOWAJordan River Scenarios. Center for Environmental Systems Research, Kurt-Wolters Strasse, p. 3.

Pak, M.V., Brieva, D.C., 2010. Designing and implementing a role-playing game: atool to explain factors, decision making and landscape transformation. Environ.Model. Softw. 25 (11), 1322e1333.

Parker, D.C., Berger, T., Manson, S.M., Use, L., 2002. Agent-based Models of Land-useand Land-cover Change: Report and Review of an International Workshop,October 4-7, 2001. LUCC Focus 1 Office, Irvine, California, USA.

Parker, D.C., Entwisle, B., Rindfuss, R.R., Vanwey, L.K., Manson, S.M., Moran, E., An, L.,Deadman, P., Evans, T.P., Linderman, M., 2008. Case studies, cross-site com-parisons, and the challenge of generalization: comparing agent-based modelsof land-use change in frontier regions. J. Land Use Sci. 3 (1), 41e72.

Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M.J., Deadman, P., 2003. Multi-agent systems for the simulation of land-use and land-cover change: a review.Ann. Assoc. Am. Geogr. 93 (2), 314e337.

Patel, M., Kok, K., Rothman, D.S., 2007. Participatory scenario construction in landuse analysis: an insight into the experiences created by stakeholder involve-ment in the Northern Mediterranean. Land Use Policy 24 (3), 546e561.

Pearl, J., 1985. Bayesian networks: a model of self-activated memory for evidentialreasoning. In: 7th Conference of the Cognitive Science Society. Computer Sci-ence Department. University of California, Irvine, 329-324.

Promper, C., Puissant, A., Malet, J.-P., Glade, T., 2014. Analysis of land cover changesin the past and the future as contribution to landslide risk scenarios. Appl.Geogr. 53, 11e19.

Rosato, P., Giupponi, C., 2003. What future for Mediterranean agriculture? A pro-posal to integrate socio-economics in climate change scenarios, Climate Changein the Mediterranean. In: Socio-economic Perspectives of Impacts, Vulnerabilityand Adaptation. Edward Elgar Publishing Ltd, Cheltenham, UK, pp. 133e158.

Saaty, T.L., 1980. The Analytic Hierarchy Process: Planning, Priority Setting, Re-sources Allocation. Mcgraw-Hill McLean, Virginia.

Saaty, T.L., Rogers, P.C., 1976. Higher education in the United States (1985e2000):scenario construction using a hierarchical framework with eigenvectorweighting. Socio Econ. Plan. Sci. 10 (6), 251e263.

Sayer, J., Sunderland, T., Ghazoul, J., Pfund, J.-L., Sheil, D., Meijaard, E., Venter, M.,Boedhihartono, A.K., Day, M., Garcia, C., 2013. Ten principles for a landscapeapproach to reconciling agriculture, conservation, and other competing landuses. Proc. Natl. Acad. Sci. 110 (21), 8349e8356.

Schmitt Olabisi, L.K., Kapuscinski, A.R., Johnson, K.A., Reich, P.B., Stenquist, B.,Draeger, K.J., 2010. Using scenario visioning and participatory system dynamicsmodeling to investigate the future: lessons fromMinnesota 2050. Sustainability2 (8), 2686e2706.

Seppelt, R., Dormann, C.F., Eppink, F.V., Lautenbach, S., Schmidt, S., 2011.A quantitative review of ecosystem service studies: approaches, shortcomingsand the road ahead. J. Appl. Ecol. 48 (3), 630e636.

Shaftel, F.R., Shaftel, G.A., 1967. Role-playing for Social Values. JohnWiley, New York.Sleeter, B.M., Sohl, T.L., Bouchard, M.A., Reker, R.R., Soulard, C.E., Acevedo, W.,

Griffith, G.E., Sleeter, R.R., Auch, R.F., Sayler, K.L., 2012. Scenarios of land use andland cover change in the conterminous United States: utilizing the specialreport on emission scenarios at ecoregional scales. Glob. Environ. Change 22 (4),896e914.

Sohl, T.L., Sayler, K.L., Bouchard, M.A., Reker, R.R., Friesz, A.M., Bennett, S.L.,Sleeter, B.M., Sleeter, R.R., Wilson, T., Soulard, C., 2014. Spatially explicitmodeling of 1992-2100 land cover and forest stand age for the conterminousUnited States. Ecol. Appl. 24 (5), 1015e1036.

Souch�ere, V., Millair, L., Echeverria, J., Bousquet, F., Le Page, C., Etienne, M., 2010. Co-constructing with stakeholders a role-playing game to initiate collective

V.R. Mallampalli et al. / Environmental Modelling & Software 82 (2016) 7e2020

management of erosive runoff risks at the watershed scale. Environ. Model.Softw. 25 (11), 1359e1370.

Sterman, J.D., 2000. Business Dynamics: Systems Thinking and Modeling for aComplex World. McGraw-Hill, Irwin, Boston.

Sterman, J.D., 2012. Sustaining Sustainability: Creating a Systems Science in aFragmented Academy and Polarized World, Sustainability Science. Springer,New York, pp. 21e58.

Strager, M.P., Rosenberger, R.S., 2006. Incorporating stakeholder preferences forland conservation: weights and measures in spatial MCA. Ecol. Econ. 58 (1),79e92.

Swart, R.J., Raskin, P., Robinson, J., 2004. The problem of the future: sustainabilityscience and scenario analysis. Glob. Environ. Change 14 (2), 137e146.

Thompson, J.R., Wiek, A., Swanson, F.J., Carpenter, S.R., Fresco, N., Hollingsworth, T.,Spies, T.A., Foster, D.R., 2012. Scenario studies as a synthetic and integrativeresearch activity for long-term ecological research. BioScience 62 (4), 367e376.

Uusitalo, L., 2007. Advantages and challenges of Bayesian networks in environ-mental modelling. Ecol. Model. 203 (3), 312e318.

van Vliet, M., Kok, K., Veldkamp, T., 2010. Linking stakeholders and modellers inscenario studies: the use of fuzzy cognitive maps as a communication andlearning tool. Futures 42 (1), 1e14.

Verburg, P.H., Overmars, K.P., 2009. Combining top-down and bottom-up dynamicsin land use modeling: exploring the future of abandoned farmlands in Europewith the Dyna-CLUE model. Landsc. Ecol. 24 (9), 1167e1181.

Voinov, A., Bousquet, F., 2010. Modelling with stakeholders. Environ. Model. Softw.25 (11), 1268e1281.

Volkery, A., Ribeiro, T., Henrichs, T., Hoogeveen, Y., 2008. Your vision or my model?Lessons from participatory land use scenario development on a European scale.Syst. Pract. Action Res. 21 (6), 459e477.

Washington-Ottombre, C., Pijanowski, B., Campbell, D., Olson, J., Maitima, J.,Musili, A., Kibaki, T., Kaburu, H., Hayombe, P., Owango, E., 2010. Using a role-playing game to inform the development of land-use models for the study ofa complex socio-ecological system. Agric. Syst. 103 (3), 117e126.

Welp, M., de la Vega-Leinert, A., Stoll-Kleemann, S., Jaeger, C.C., 2006. Science-basedstakeholder dialogues: theories and tools. Glob. Environ. Change 16 (2),170e181.

Yu, W., Zang, S., Wu, C., Liu, W., Na, X., 2011. Analyzing and modeling land use landcover change (LUCC) in the Daqing City, China. Appl. Geogr. 31 (2), 600e608.

Zadeh, L.A., 1965. Fuzzy sets. Inf. Control 8 (3), 338e353.Zimmermann, H.J., 2010. Fuzzy set theory. Wiley Interdiscip. Rev. Comput. Stat. 2

(3), 317e332.